Intel's new tool lets you benchmark DLSS, FSR, and XeSS image quality

When it comes to modern GPUs, raw performance remains a key metric for determining whether you can game in 1080p, 1440p, or 4K. However, in the age of Super Resolution and Frame Generation technologies, including NVIDIA DLSS, AMD FSR, and Intel XeSS, raw performance is only a part of the overall picture and story.

2

In fact, without DLSS, real-time Path Tracing in games like DOOM: The Dark Ages and Half-Life 2 RTX runs slow and looks noticeably worse, even on a GeForce RTX 5080. With new AI-enhanced rendering technologies designed to boost image quality, such as Neural Radiance Cache and Ray Reconstruction, overall image quality and presentation are now key metrics for measuring a GPU's performance and the experiences it can deliver. NVIDIA has already confirmed that when it comes to its latest generation of GeForce RTX 50 Series GPUs, over 90% of users are gaming with DLSS enabled.

This is where Intel's latest tool, the Computer Graphics Visual Quality Metric (CGVQM), enters the picture. Although it evaluates gameplay video, it's a tool built for the new era of DLSS, FSR, and XeSS-powered gaming.

"Whether it's fine-tuning a game engine, streaming a high-performance game over the cloud, or evaluating a new rendering technique, video quality assessment plays a crucial role across a range of scenarios," Intel writes. "This is where objective video quality metrics come in. These automated tools act as proxies for human perception, predicting how good (or bad) a video looks compared to a pristine reference. Computer Graphics Video Quality Metric (CGVQM) [is] designed to capture complex artifacts introduced by modern rendering techniques in computer graphics."

Intel's open-source solution is capable of detecting "ghosting, temporal flicker, shimmering noise, and even hallucinated textures introduced by neural networks." As a tool for evaluating image quality, it can be used by developers to fine-tune image quality settings or to aid in training neural networks. The tool is still in development, as it's based on evaluating video files of gameplay, it requires a "perfect, undistorted version" for accurate results.

It's a fascinating tool, and future updates will introduce new capabilities for testing the "realism" of AI-generated pixels as opposed to their accuracy. You can check out the full findings here, with Intel noting that it will release the full research paper and source code soon.